DualPG-DTA: A Large Language Model-Powered Graph Neural Network Framework for Enhanced Drug-Target Affinity...

Area of Science:

• Computational chemistry and drug discovery
• Bioinformatics and machine learning
• Pharmacology and oncology

Background:

• Accurate prediction of drug-target interactions is essential for efficient drug discovery.
• Existing models often lack the precision required for complex molecular interactions.
• Developing novel therapeutics for drug-resistant cancers remains a significant challenge.

Purpose of the Study:

• To introduce DualPG-DTA, a general framework for predicting drug-target binding affinity.
• To validate the framework's performance against existing state-of-the-art models.
• To identify and characterize novel therapeutic agents for acute myeloid leukemia (AML).

Main Methods:

• Integration of two pre-trained language models for atomic and residue-level molecular representations.
• Construction of dual molecular-protein graphs processed by graph neural networks with dynamic attention.
• Fusion of context-aware sequence-level features via a multimodal module for affinity prediction.

Main Results:

• DualPG-DTA demonstrated superior performance across all metrics compared to existing benchmark models.
• Compound C1 was identified as a novel CDK9 inhibitor with high affinity (IC50 = 1.2 nM) and selectivity.
• Compound C1 exhibited potent antitumor efficacy in a Venetoclax-resistant AML model with favorable pharmacokinetic and safety profiles.

Conclusions:

• DualPG-DTA provides a unified and accurate framework for binding affinity prediction.
• Compound C1 represents a promising therapeutic lead for overcoming Venetoclax resistance in AML.
• The study highlights the potential of integrating advanced AI techniques in drug discovery and development.